A train online detection method

Abstract

The present application relates to the field of visual detection, and discloses a train online detection method. The method adopts a mode that 3D distance measurement of a laser radar is mainly used and 2D image recognition of a third camera is secondarily used for judging whether a region with different characteristics on the appearance of the train changes. The 3D distance measurement comprises the following steps: S11. obtaining a three-dimensional point data set of the appearance of the train by using the laser radar; S12. dividing the three-dimensional point data set into a plurality of columns of first data blocks i in a first interval along a scanning direction; S13. dividing each column of first data blocks i into a plurality of rows of second data blocks ij in a second interval along the vertical direction of the scanning direction; S14. respectively performing distance distribution statistics of three-dimensional points on each second data block ij; and S15. if the change amplitude of the distance distribution of the three-dimensional points in the first time and the second time is less than a threshold value, then the region of the appearance of the train corresponding to the second data block ij has no change. The present application is used to solve the problem of low accuracy of abnormal situation recognition in train appearance detection.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of visual detection, more particularly, to a train online detection method. BACKGROUND

[0002] As a common means of transportation in modern society, trains provide convenience for people's travel and play an important role in social development. As a common means of transportation, the safe operation of trains is of great importance, so trains need to be routinely safety detected every trip or every day. The routine safety detection procedure is complicated and the detection frequency is high, and automatic safety detection is the focus of research in the relevant field.

[0003] Trains have a relatively complex structure and many key components (pantograph, bogie), and the train's running process may cause abnormalities of the train body due to various conditions such as environment, part aging, human operation, etc., such as foreign matter falling on the roof, missing of key components, deformation or change of component position, etc.; and the interference of foreign matter, the missing of key components and the change of position will affect the safe operation of the train. Therefore, in the routine safety detection, the inspection of the train appearance is an essential important link. In addition, the train generates a lot of long-time vibration during movement, which makes the fasteners on the train prone to loosen, and if not found in time, it is easy to cause serious accidents. Therefore, the looseness detection of the fasteners is also an essential important link and the most numerous and frequent detection.

[0004] In the prior art, a common train appearance detection method is to use a camera to shoot the passing train to obtain an image, and to identify the image by manual or computer software. When manual detection is used, the abnormal features of the train body are easily missed due to visual fatigue or incomplete patterns; when automatic recognition by computer software is used, high-standard image shooting requirements and accurate recognition algorithms need to be met to identify the abnormal features of the train body, and complex on-site environments and train appearances often lead to image recognition errors and cannot give accurate detection results.

[0005] Laser radar is a sensor that uses laser light to achieve precise ranging. Laser radar emits laser pulses that are reflected back when they encounter surrounding objects. By measuring the time it takes for the laser to reach each object and return, the precise distance of the object can be calculated. Laser radar emits thousands of pulses per second, and by collecting these distance measurements (distance points), a model of the appearance of the scanned object can be constructed. SUMMARY

[0006] The present application aims to overcome at least one of the deficiencies of the prior art, and provides a train online detection method to solve the problem of low accuracy of abnormal situation recognition in train appearance detection.

[0007] The technical scheme adopted by the present application is a train online detection method, for the region with variable shape position relationship or light transmission characteristics on the train appearance, a third camera is used to obtain a third image of the surface, then an image recognition module is used to recognize the features in the third image, finally an analysis module is used to compare the features of the third images obtained in the first and second times, to determine whether the region has changed;

[0008] For the region with fixed shape position relationship or non-light transmission on the train appearance, a laser radar is used for three-dimensional scanning detection, including the following steps:

[0009] S11. Scanning the train appearance using the laser radar, and obtaining a three-dimensional point data set of the train appearance;

[0010] S12. Dividing the three-dimensional point data set into a plurality of columns of first data blocks i (i is a positive integer) along the scanning direction and at a first interval;

[0011] S13. Dividing each column of first data blocks i into a plurality of rows of second data blocks ij (i, j are positive integers) along the vertical direction of the scanning direction and at a second interval;

[0012] S14. Statistically analyzing the distance distribution of three-dimensional points for each second data block ij;

[0013] S15. A plurality of second data blocks Aij are obtained by first scanning, and a plurality of second data blocks Bij are obtained by second scanning, if the change amplitude of the three-dimensional point distance distribution of the second data block Bij and the second data block Aij is less than a threshold value, then the region of the train appearance corresponding to the second data block ij has no change.

[0014] The scheme adopts the combination of camera and laser radar to detect the abnormal appearance of the train. On the one hand, for the region with variable shape position relationship or light transmission characteristics, such as the windscreen at the connection end of the train carriage and the glass window on the side of the train, a third camera is used to shoot a 2D image for feature recognition. The appearance of the windscreen is scalable and has no fixed shape. For accurate 3D distance measurement, the data measured each time is different, and it is impossible to distinguish between normal and abnormal conditions; while the 2D image can be accurately recognized by gray scale, AI shape and other algorithms. If the recognized features of the first and second times change, the region has changed, thereby discovering the abnormal conditions such as damage, shedding and foreign matter attachment of the windscreen. In addition, the glass window has light transmission characteristics, and the general laser radar cannot realize 3D distance measurement on its surface, and can only be recognized and judged by 2D image, thereby discovering the abnormal conditions of the glass window. On the other hand, for the regions other than this, including the regions with fixed shape position relationship or non-light transmission, laser radar is used for 3D distance measurement and recognition. For most regions of the train, or key components such as the running part and the pantograph, 3D distance measurement can be applied. The scheme uses 3D distance measurement of laser radar as the main method and 2D image recognition of the third camera as the auxiliary method, so that it can exert its respective performance for different regions of the train appearance, reduce the false alarm frequency of a small number of special components under the premise of accurate recognition of abnormal conditions of most key components, and thereby improve the overall accuracy of abnormal condition recognition.

[0015] In the scheme, the 2D image recognition of the third camera can refer to the prior art. Only the 3D distance measurement of the laser radar is described in detail below.

[0016] In the scheme, the laser radar is used to scan the appearance of the train, obtain three-dimensional point data, and further recognize abnormal conditions based on the three-dimensional point data. Since the three-dimensional point data is distance data, compared with the traditional image recognition technology, the color change, pattern change and surface stain attachment of the train appearance, as well as the on-site lighting conditions, will not interfere with the laser radar to obtain the three-dimensional point data of the train appearance, thereby ensuring the accuracy of abnormal condition recognition from the perspective of data source.

[0017] Further, the scheme realizes the recognition of abnormal conditions on the train appearance by performing distance value distribution statistics on three-dimensional points in a certain spatial range and comparing the distance distributions of three-dimensional points before and after. In a certain spatial range, if structural deformation, loss and relative position change of parts on the train appearance, and surface attachment of foreign matter with a certain volume occur, the distance value of the three-dimensional points obtained by scanning the position will change, and the distance distribution of the three-dimensional points in the spatial range will also change accordingly. That is, the distance distribution of three-dimensional points corresponds to the train appearance, and the change of the distance distribution of three-dimensional points indicates that the train appearance has an abnormal condition. The distance distribution of three-dimensional points can be a quantity distribution or a probability distribution. Compared with the traditional image recognition technology, the scheme does not need to perform feature recognition on the three-dimensional point data set first, and then compare and judge the features before and after. Instead, it directly compares the distance distribution of three-dimensional points in a certain spatial range, and then determines whether an abnormal condition occurs in the spatial range. Since the scheme only performs statistical and comparison calculation of three-dimensional points, the operation and judgment process is simplified, the interference factors are reduced, and the speed and accuracy of abnormal condition recognition are improved from the perspective of judgment method. In addition, in the case of considering the interference of other external factors, when the change amplitude of the distance distribution of three-dimensional points is greater than a threshold value, it indicates that an abnormal condition occurs on the train appearance.

[0018] Secondly, the scheme divides the scanned three-dimensional point data set into a plurality of second data blocks ij by grid cutting (first interval and second interval), and then performs distance distribution statistics of three-dimensional points. On the one hand, the distance distribution of three-dimensional points can only determine whether an abnormal condition occurs in a certain spatial range, and cannot give the specific position; the grid division of the overall appearance of the train can confirm the occurrence area of the abnormal condition. Taking the scanning start end of the train as the reference, the occurrence area of the abnormal condition on the train appearance can be located by the number ij of the second data block ij, the first interval and the second interval. On the other hand, based on the distance distribution statistics of three-dimensional points, the smaller the statistical spatial range, the higher the sensitivity to changes and the higher the resolution of foreign matter detection. For example, a 20mm×10mm foreign matter accounts for 0.01% in a 2000mm×10000mm range, and accounts for 1% in a 200mm×100mm range. Obviously, when the same foreign matter changes occur, the change amplitude of the distance distribution of three-dimensional points in the 200mm×100mm range is larger, and the abnormal condition is easier to be found. Therefore, the second data block ij with a smaller range can improve the sensitivity or resolution of abnormal condition recognition.

[0019] Preferably, before step S14, the first data block i or the second data block ij is subjected to interpolation or fitting processing of three-dimensional points to make the three-dimensional point density the same. In step S11, when the train is scanned by the laser radar, due to the difficulty in keeping the relative motion speed unchanged, the three-dimensional point density obtained at different positions along the scanning direction is different, and the three-dimensional point density obtained at the same position twice is also different. The three-dimensional point density of the second data block ij is the same, so that the standard of distance value distribution statistics can be unified, which is beneficial to improve the accuracy of the obtained three-dimensional point distance distribution and further improve the accuracy of the abnormal situation recognition. Secondly, the scheme uses a set fixed value to interpolate the points at positions with sparse three-dimensional point density and to fit the points at positions with dense three-dimensional point density. The interpolation or fitting processing can be performed after step S11 or step S12 or step S13.

[0020] In the scheme, according to the relative motion relationship between the laser radar and the train, there can be various scanning modes. It can be that the train is fixed, and the 2D laser radar or the 3D laser radar scans on the surface of the train. It can also be that the 2D laser radar or the 3D laser radar is fixed, and the train passes through the scanning area thereof.

[0021] Preferably, in step S11, the train travels on the track, and the laser radar is fixed outside the track; the train passes through the scanning area of the laser radar, so that a three-dimensional point data set of the appearance of the train is obtained. In the mode that the laser radar is fixed and the train moves, the motion part of the laser radar can be saved, and online detection of the train can be realized.

[0022] In the scheme, the three-dimensional point data set obtained in step S11 can be a three-dimensional model of the appearance of the train constructed, and steps S12 to S15 are for statistical judgment of the three-dimensional points of the three-dimensional model. The laser radar scans the train, and transmits the original scanned frame data and speed or position data to a modeling module, so as to realize construction of the three-dimensional model.

[0023] Preferably, the three-dimensional point data set obtained in step S11 can also be original scanned frame data, and steps S12 to S15 directly perform statistical judgment on the three-dimensional points of the frame data. Since the modeling process of the appearance of the train is cancelled, the acquisition speed of the three-dimensional point data set is accelerated and the data interference factors are reduced, which is beneficial to further improve the speed and accuracy of the abnormal situation recognition.

[0024] When the frame data is adopted, the real length corresponding to the three-dimensional point data cannot be obtained due to the lack of reference coordinates in the frame data itself. Therefore, the specific embodiment further comprises a sensor, and a set of time stamp sets containing the time T i (i is a positive integer) corresponding to each first interval distance of the train advancing is obtained by using the sensor to trigger when the train passes through the scanning area of the laser radar. The length of the three-dimensional point data between any two adjacent time stamps is the first interval. Then, in step S12, the frame data is segmented by the time stamps, so that the three-dimensional point data set is divided into a plurality of first data blocks i according to the first interval.

[0025] Further, the sensor is a wheel sensor, and a plurality of wheel sensors are arranged on the track in sequence; the plurality of wheel sensors are used to generate a set of trigger signals containing the front and rear sequence when the train passes through. The plurality of wheel sensors are arranged on the track in sequence according to the first interval, and when the wheels of the train pass through the wheel sensors in sequence, the trigger signals generated by the plurality of wheel sensors contain the information of the first interval. The set of trigger signals is recorded by a clock, and a set of time stamp sets required above can be obtained.

[0026] Preferably, the first interval or the second interval is 300mm to 900mm. When the three-dimensional point data set is a three-dimensional model, the first interval and the second interval can be selected within the above range according to the detection position of the train appearance, requirements, and the operation processing capacity of the equipment. The smaller the first interval and the second interval, the greater the amount of operation in the detection process of meshing and judgment, and the higher the sensitivity or resolution of the abnormal situation identification. When the three-dimensional point data set is frame data, the first interval is selected within the above range, which can effectively avoid the scanning distortion caused by the fluctuation of the train speed, and further ensure the accuracy of the abnormal situation identification from the data source perspective. The speed of the train inevitably fluctuates when passing through the scanning of the laser radar, but due to the inertia of the train and the limitation of the acceleration and deceleration size, the speed of the train in a small range basically remains unchanged. That is, the train driving in a small range can be regarded as uniform motion. Through repeated experimental tests, it is found that the range of the first interval should be between 300mm and 900mm.

[0027] Preferably, the scanning frequency of the laser radar is not less than 500HZ.

[0028] Preferably, the plurality of laser radars are arranged at the upper left corner, the upper right corner, the lower left corner, the lower right corner and the front of the train section respectively, and the scanning area of the laser radar covers the roof and the running part of the train. Among them, the laser radars at the upper left corner and the upper right corner respectively complete the appearance scanning of the train roof from the left and right sides, including the pantograph, refrigeration equipment and other key components; the laser radars at the lower left corner, the lower right corner and the front respectively complete the appearance scanning of the train bottom from the left and right sides and the front, including the bogie and other key components.

[0029] Optionally, a second camera is arranged side by side on one side of the laser radar, and the second camera is used to synchronously obtain a set of image groups of the appearance of the train when the train passes.

[0030] Optionally, the method further comprises step S16. If the appearance of the train region corresponding to the second data block ij changes, an alarm message is sent to the user, and an image of the corresponding position is pushed for manual confirmation.

[0031] In this scheme, the second camera is used as a supplement to the laser radar to realize artificial auxiliary diagnosis. The second camera is arranged side by side in parallel on one side of the laser radar, and the visual angle range of the second camera covers the train appearance scanned by the laser radar. The second camera can use the above-mentioned timestamp as a mark for confirming the position of the shooting part. When an abnormal situation is identified in the region corresponding to a certain second data block ij, the corresponding timestamp is found through the number i, and then the image involving the first data block i region is found from the image group through the timestamp, and finally the image is sent to the user for secondary confirmation of whether an abnormality has occurred.

[0032] The laser radar scheme of the present application cannot realize the detection of fastener loosening, and the present application further provides a fastener loosening detection method based on 2D image recognition technology, comprising the following steps:

[0033] S21. A first mark line and a second mark line are arranged on the end face of the fastener;

[0034] S22. When initialized, a first image of the end face of the fastener is obtained using the first camera, and the first image is transmitted to the image recognition module;

[0035] S23. The first mark line and the second mark line in the first image are recognized using the image recognition module, and the reference slope K1 of the first mark line and the reference slope K2 of the second mark line are obtained;

[0036] S24. When detected, a second image of the end face of the fastener is obtained using the first camera, and the second image is transmitted to the image recognition module;

[0037] S25. The first mark line and the second mark line in the second image are recognized using the image recognition module, and the actual slope K5 of the first mark line and the actual slope K6 of the second mark line are obtained;

[0038] S26. The slope change of the first mark line and the second mark line is judged using the analysis module, and the loosening result of the fastener is given.

[0039] The scheme takes the fastener loosening detection applied to the bogie of the train as an example. A first camera is fixed on one side of the track to form a certain detection area within the visual angle range of the first camera. A wheel sensor is arranged on the track, and the wheel sensor is used to trigger a shooting signal to the first camera. When the train passes through the detection area at a certain speed, the wheel sensor detects the wheels of the train and sends a shooting signal to the first camera, the first camera shoots an image of the bogie containing a plurality of fasteners, and an image recognition module and an analysis module process the image respectively to obtain the loosening condition of each fastener.

[0040] In the scheme, after the fastener is fastened, a maintenance personnel sets a first mark line and a second mark line on the end face of the fastener. When the train enters the detection area for the first time (at the initialization time), the first camera triggers shooting at a fixed point to obtain a first image of the bogie, and the first image covers the end face of the fastener; the image recognition module obtains a reference slope K1 of the first mark line and a reference slope K2 of the second mark line. When the train enters the detection area for the second time (at the detection time), the first camera triggers shooting at a fixed point to obtain a second image of the bogie, and the second image covers the end face of the fastener; the image recognition module obtains an actual slope K5 of the first mark line and an actual slope K6 of the second mark line. Then, the analysis module compares the difference between the reference slope and the actual slope with a preset threshold value, if the difference is greater than the threshold value, it indicates that the fastener has loosened; if the difference is less than the threshold value, it indicates that the fastener remains fastened. And there are two ways to calculate the difference between the reference slope and the actual slope, one is to first calculate a first difference between the reference slope K1 and the actual slope K5 of the first mark line, and a second difference between the reference slope K2 and the actual slope K6 of the second mark line, and then average the first difference and the second difference; the other is to first calculate the slope of the angle bisector of the first mark line and the second mark line in the first image as the reference slope, and then calculate the slope of the angle bisector of the first mark line and the second mark line in the second image as the actual slope, and finally calculate the difference between the reference slope and the actual slope of the angle bisector.

[0041] In the scheme, on the one hand, since the first image and the second image obtained by the first camera both include the first mark line and the second mark line, when evaluating the slope change, the analysis module eliminates the error caused by image recognition by solving the average, thereby solving the problem of low accuracy when applying visual detection to fastener loosening. On the other hand, since the double insurance mode of two mark lines is adopted, in the case of image recognition error of one of the mark lines, the analysis module can still determine the loosening result of the fastener through the slope change of the other mark line, thereby solving the problem of high false positive rate when applying visual detection to fastener loosening.

[0042] It should be noted that the first and second time of the train entering the detection area only indicates the order, and does not represent the actual number of times of entering the train. That is, when the train enters the detection area next time, the last time detection result can be used as initialization data, the last image as the first image, the actual slope of the first and second mark lines as the reference slope K1 and the reference slope K2, and the next slope change analysis is performed. Secondly, the first and second mark lines are not limited to a single line segment, but can also be the edge of a figure. For example, the first and second mark lines are two sides of a triangle. Furthermore, according to the angle range of the first camera, the first and second images can cover multiple fasteners, and the image analysis module numbers the multiple fasteners according to the rules, and the image recognition module and the analysis module can also process the first and second mark lines of multiple fasteners in an image at the same time.

[0043] Preferably, step S23 further comprises: if the image recognition module does not obtain the reference slope K1 of the first mark line and the reference slope K2 of the second mark line, an alarm information is sent to the user, and the step S22 is returned. The added judgment step can ensure that the reference slope K1 of the first mark line and the reference slope K2 of the second mark line are obtained during initialization, so that the loosening of the fastener is judged by solving the average during detection.

[0044] Preferably, it further comprises step S27: if the slope change of the first and second mark lines cannot be identified or exceeds the threshold value, an alarm information is sent to the user, and the second image is pushed for manual confirmation at the same time.

[0045] Preferably, the first mark line and the second mark line intersect at a point on the end surface of the fastener. After the intersection of the first and second mark lines, the first and second mark lines are associated with each other due to the appearance of the intersection point. When the image recognition module identifies the first and second mark lines, the intersection point is added as an identification feature, which is beneficial to reduce the identification error of the image recognition module and improve the accuracy of the obtained reference slope K1 and reference slope K2.

[0046] Further, it further comprises a third mark line, the third mark line intersects with the first mark line and the second mark line respectively; the third mark line is used to correct the slope change of the first and second mark lines. The addition of the third mark line makes the first and second mark lines respectively add another intersection point. When the image recognition module identifies the first and second mark lines, two intersection points are added as identification features, which further reduces the identification error of the image recognition module and improves the accuracy of the obtained reference slope K1 and reference slope K2.

[0047] Further, the first mark line and the second mark line are cross lines located at the center of the end face of the fastener. When being cross lines, the included angle of the first mark line and the second mark line is maximum, and the pixel interference between the first mark line and the second mark line is minimum near the intersection point. In addition, the first mark line and the second mark line cover the largest range on the end face of the fastener, and even if the first mark line and the second mark line are partially damaged or partially blocked, it will not affect the image recognition module. Therefore, the first mark line and the second mark line in the shape of cross lines further improve the recognition accuracy of the image recognition module.

[0048] In the present scheme, the setting of the first mark line and the second mark line on the end face of the fastener includes two ways, one is based on the traditional marking line method, and the other is based on the pasting method of mark sticker.

[0049] When using the marking line method, step S21 includes using a marking pen to draw the first mark line and the second mark line on the end face of the fastener. Since the maintenance personnel will inevitably set the check line after completing the fastening of the fastener, and it is required to add the drawing of the first mark line and the second mark line, it will not bring additional application cost, thereby reducing the influence on the existing safety detection process, making the method easy to popularize and use. In addition, the image recognition module uses a regression algorithm to calculate the slope of the first mark line and the second mark line, and the first mark line and the second mark line do not need to be straight line segments, thereby reducing the drawing requirements for the maintenance personnel.

[0050] Further, the first mark line and / or the second mark line extend from the end face of the fastener to the surface of the object being fastened. The first mark line or the second mark line on the end face of the fastener is used for the judgment of the recognition and analysis module of the image recognition module; the first mark line or the second mark line on the extended part is equivalent to the traditional check line, which is used for manual detection. At this time, it is equivalent to that the maintenance personnel directly draw the first mark line and the second mark line on the basis of the original check line. The extended first mark line or second mark line not only adapts to visual detection and ensures the reliability of image recognition, but also plays the function of automatic loosening detection; and it also retains the function of the original check line, which can be directly compared and checked, realizes man-machine integration, and realizes flexible switching of automatic / manual detection.

[0051] When using the pasting method, step S21 includes pasting the mark sticker on which the first mark line and the second mark line are drawn to the end face of the fastener. The first mark line and the second mark line are batch printed on the mark sticker, and the thickness, straightness, color depth, etc. of the first mark line and the second mark line can be uniformly guaranteed, thereby reducing the recognition difficulty of the image recognition module and improving the recognition accuracy. When used, the maintenance personnel can paste the mark sticker on the end face of the fastener after fastening the fastener.

[0052] Further, the first mark line and the second mark line have waterproof properties, reflective properties, and / or waterproof properties and fluorescent properties. The waterproof properties are to prevent the first mark line and the second mark line from being easily damaged, and the reflective properties and the fluorescent properties are to improve the exposure of the first mark line and the second mark line in the first image or the second image, improve the contrast, and make the image recognition module easier and more accurate to recognize.

[0053] Compared with the prior art, the present application has the following advantages:

[0054] The present application uses a laser radar to scan the appearance of the train, avoids color changes, pattern changes, and surface stains of the train appearance, and interference caused by on-site lighting conditions, and guarantees the accuracy of abnormal situation recognition from the perspective of data sources. The present application simplifies the operation and judgment process by comparing the three-dimensional point distance distribution within a certain spatial range, reduces interference factors, and further improves the speed and accuracy of abnormal situation recognition from the perspective of judgment method. The present application improves the sensitivity or resolution of abnormal situation recognition by grid processing the three-dimensional point data set obtained by scanning. The present application obtains a timestamp set by triggering the sensor, and then directly uses the original scanning frame data as the three-dimensional point data set for statistical judgment, further improving the speed and accuracy of abnormal situation recognition.

[0055] The present application uses 3D distance measurement of a laser radar as the main method, and 2D image recognition of a third camera as the auxiliary method, so that it can exert its respective performance on different areas of the train appearance. On the premise of ensuring the accurate recognition of abnormal situations of most key components, the false alarm frequency of a small number of special components is reduced, and the overall abnormal situation recognition accuracy is improved.

[0056] In the present application, on the one hand, since the first mark line and the second mark line are included in the first image and the second image obtained by the first camera, when evaluating the slope change, the analysis module eliminates the error caused by image recognition by solving the average, and further solves the problem of low accuracy when applying visual detection to loosen fasteners. On the other hand, since the double insurance method of two mark lines is adopted, in the case of image recognition error of one of the mark lines, the analysis module can still determine the looseness of the fastener through the slope change of the other mark line, and further solves the problem of high false alarm rate when applying visual detection to loosen fasteners. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 FIG. 1 is a top view of an embodiment of the present application.

[0058] Figure 2 FIG. 1 is a train cross-sectional view of an embodiment of the present application.

[0059] Figure 3A schematic diagram of a three-dimensional point data set grid processing of embodiment 1 of the present application.

[0060] Figure 4 A schematic diagram of the second data block A11 of embodiment 1 of the present application.

[0061] Figure 5 A three-dimensional point distance distribution diagram of the second data block A11 of embodiment 1 of the present application.

[0062] Figure 6 A schematic diagram of the second data block B11 of embodiment 1 of the present application.

[0063] Figure 7 A three-dimensional point distance distribution diagram of the second data block B11 of embodiment 1 of the present application.

[0064] Figure 8 A schematic diagram of the principle of timestamp division frame data of embodiment 1 of the present application.

[0065] Figure 9 A partial view of the first image at initialization in embodiment 2 of the present application.

[0066] Figure 10 A partial view of the second image after loosening of the fastener in embodiment 2 of the present application.

[0067] Figure 11 A partial view of the first image at initialization in embodiment 3 of the present application.

[0068] Figure 12 A partial view of the second image after loosening of the fastener in embodiment 3 of the present application.

[0069] Figure 13 A partial view of the first image at initialization in embodiment 4 of the present application.

[0070] Figure 14 A partial view of the second image after loosening of the fastener in embodiment 4 of the present application.

[0071] Label explanation: pantograph 11, bogie 12, fastener 13, first mark line 21, second mark line 22, third mark line 23, anti-loose line 24, laser radar 30, first camera 41, second camera 42, third camera 43, wheel sensor 50. DETAILED DESCRIPTION

[0072] The drawings of the present application are only for illustrative purposes and cannot be understood as a limitation of the present application. In order to better illustrate the following embodiments, some components of the drawings may be omitted, enlarged or reduced, and do not represent the actual size of the product; it is understandable for those skilled in the art that some well-known structures and their descriptions in the drawings may be omitted.

[0073] Example 1

[0074] like Figure 1 , Figure 2 As shown, this embodiment is a train online detection method. For areas on the train exterior that have changing shape, positional relationship or light transmission characteristics, a third camera is used to acquire a third image of its surface. Then, an image recognition module is used to identify the features in the third image. Finally, an analysis module is used to compare the features of the first and second third images to determine whether the area has changed.

[0075] For areas on the train's exterior with fixed shapes and positions or that are not translucent, a three-dimensional scanning inspection is performed using LiDAR, including the following steps:

[0076] S11. Use LiDAR to scan the exterior of the train and obtain a set of three-dimensional point data of the train's exterior;

[0077] S12. Along the scanning direction, divide the three-dimensional point data set into several columns of first data blocks i (i is a positive integer) according to the first spacing;

[0078] S13. Along the vertical direction of the scanning direction, divide each column of the first data block i into several rows of second data blocks ij (i and j are positive integers) according to the second spacing;

[0079] S14. Perform distance distribution statistics on three-dimensional points for each second data block ij;

[0080] S15. The first scan obtains several second data blocks Aij, and the second scan obtains several second data blocks Bij. If the change in the three-dimensional point distance distribution between the second data block Bij and the second data block Aij is less than the threshold, then the area of ​​the train appearance corresponding to the second data block ij remains unchanged.

[0081] The scheme adopts a combination of cameras and lidar to detect train appearance abnormalities. On the one hand, for regions with changing shape position relationships or light transmission characteristics, such as train car connection end windshields and train side glass windows, a third camera is used to capture 2D images for feature recognition. The appearance of the windshield is scalable and has no fixed shape. For accurate 3D distance measurement, the data measured each time is different, and it is impossible to distinguish between normal and abnormal situations; while 2D images can be accurately identified through grayscale, AI shape, and other algorithms. If the first and second times the identified features change, the region has changed, thus discovering abnormal situations such as windshield damage, shedding, and foreign object attachment. In addition, glass windows have light transmission characteristics, and general lidar cannot achieve 3D distance measurement on their surface, and can only be identified and judged through 2D images to discover abnormal situations of glass windows. On the other hand, for regions other than this, including regions with fixed shape position relationships or non-light transmission, lidar is used for 3D distance measurement identification. For most regions of the train, or key components such as the running gear and pantograph, 3D distance measurement is applicable. This scheme uses 3D distance measurement by lidar as the main method, and 2D image recognition by the third camera as the auxiliary method, so that it can exert its respective effectiveness on different regions of the train appearance, reduce the false alarm frequency of a small number of special components under the premise of ensuring accurate identification of abnormal situations of most key components, and thus improve the overall accuracy of abnormal situation identification.

[0082] In this scheme, 2D image recognition by the third camera can refer to existing technology. The following only describes the 3D distance measurement by lidar in detail.

[0083] In this scheme, a laser radar is used to scan the train appearance, obtain its three-dimensional point data, and further identify abnormal situations based on the three-dimensional point data. Since the three-dimensional point data is distance data, compared with traditional image recognition technology, color changes, pattern changes, and surface stains on the train appearance, as well as on-site lighting conditions, will not interfere with the laser radar obtaining the three-dimensional point data of the train appearance, thus ensuring the accuracy of abnormal situation identification from the perspective of data source.

[0084] Further, the scheme realizes the recognition of abnormal conditions on the train appearance by performing distance value distribution statistics on three-dimensional points in a certain spatial range and comparing the distance distributions of three-dimensional points before and after. In a certain spatial range, if structural deformation, loss and relative position change of parts on the train appearance, and surface attachment of foreign matter with a certain volume occur, the distance value of the three-dimensional points obtained by scanning the position will change, and the distance distribution of the three-dimensional points in the spatial range will also change accordingly. That is, the distance distribution of three-dimensional points corresponds to the train appearance, and the change of the distance distribution of three-dimensional points indicates that the train appearance has an abnormal condition. The distance distribution of three-dimensional points can be a quantity distribution or a probability distribution. Compared with the traditional image recognition technology, the scheme does not need to perform feature recognition on the three-dimensional point data set and then compare and judge the features before and after. Instead, the distance distribution of three-dimensional points in a certain spatial range is directly compared, and then it is determined whether an abnormal condition occurs in the spatial range. Since the scheme only performs statistical and comparison calculation of three-dimensional points, the operation and judgment process is simplified, the interference factors are reduced, and the speed and accuracy of abnormal condition recognition are improved from the perspective of judgment method. In addition, when the change amplitude of the distance distribution of three-dimensional points is greater than a threshold value in the case of considering other external factors, it is determined that an abnormal condition occurs on the train appearance.

[0085] As shown in Figure 3 , the scheme cuts and divides the three-dimensional point data set obtained by scanning into a plurality of second data blocks ij according to gridding (first interval and second interval), and then performs distance distribution statistics of three-dimensional points. On the one hand, the distance distribution of three-dimensional points can only determine whether an abnormal condition occurs in a certain spatial range, and cannot give a specific position; the gridding division of the overall appearance of the train can confirm the occurrence area of the abnormal condition. Taking the scanning start end of the train as a reference, the train appearance area where the abnormal condition occurs can be located through the number ij of the second data block ij, the first interval and the second interval. On the other hand, based on the distance distribution statistics of three-dimensional points, the smaller the statistical spatial range, the higher the sensitivity to changes and the higher the resolution of foreign matter detection. For example, a 20mm×10mm foreign matter accounts for 0.01% in a 2000mm×10000mm range, and accounts for 1% in a 200mm×100mm range. Obviously, when the same foreign matter changes occur, the change amplitude of the distance distribution of three-dimensional points in the 200mm×100mm range is larger, and the abnormal condition is easier to be found. Therefore, the second data block ij with a smaller range can improve the sensitivity or resolution of abnormal condition recognition.

[0086] In this embodiment, as shown in Figures 4 to 7 , the above judgment process is further described taking the second data block 11 as an example. Figure 4 and Figure 6Visualization images of the three-dimensional point data sets obtained in the first and second time respectively in the local area of the train appearance, the line segments in the images are composed of several three-dimensional points with distance values. Figure 5 and Figure 7 The three-dimensional point distance distributions of the second data block 11 obtained in the first and second time respectively. When the local area has a certain volume of foreign matter (cube) attached, the three-dimensional point distance distribution of the second data block changes accordingly. By judging the change amplitude, it can be identified that the local area has an abnormal situation.

[0087] Preferably, before step S14, the three-dimensional points of the first data block i or the second data block ij are interpolated or fitted to make their three-dimensional point densities the same. In step S11, when the train is scanned by the laser radar, due to the difficulty of keeping the relative motion speed unchanged, the three-dimensional point densities obtained at different positions along the scanning direction are different, and the three-dimensional point densities obtained at the same position in the first and second time are also different. The three-dimensional point densities of the second data block ij are the same, which can unify the statistical standard of distance value distribution and is beneficial to improve the accuracy of the obtained three-dimensional point distance distribution and further improve the accuracy of abnormal situation identification. Secondly, the scheme uses a set fixed value to interpolate the points at positions with sparse three-dimensional point density and to fit the points at positions with dense three-dimensional point density. The interpolation or fitting processing can be performed after step S11 or step S12 or step S13.

[0088] Preferably, in step S11, the train travels on the track, and the laser radar is fixed outside the track; the train passes through the scanning area of the laser radar, thereby obtaining the three-dimensional point data set of the train appearance. The laser radar fixed and the train moving can save the motion part of the laser radar and realize online detection of the train.

[0089] In the scheme, the three-dimensional point data set obtained in step S11 can be a three-dimensional model of the train appearance constructed, and steps S12 to S15 are statistical judgments on the three-dimensional points of the three-dimensional model. The laser radar scans the train and transmits the original scanned frame data and speed or position data to the modeling module, thereby realizing the construction of the three-dimensional model.

[0090] Preferably, the three-dimensional point data set obtained in step S11 can also be the original scanned frame data, and steps S12 to S15 directly perform statistical judgments on the three-dimensional points of the frame data. Since the modeling process of the train appearance is cancelled, the acquisition speed of the three-dimensional point data set is accelerated and the data interference factors are reduced, which is beneficial to further improve the speed and accuracy of abnormal situation identification.

[0091] As Figure 8As shown, when frame data is used, the real length corresponding to the three-dimensional point data cannot be obtained due to the lack of reference coordinates in the frame data itself. Therefore, the specific embodiment further includes a sensor, which is used to trigger a set of time stamps containing the time T i (i is a positive integer) corresponding to each first interval distance of the train advancement while the train passes through the scanning area of the laser radar. The length of the three-dimensional point data between any two adjacent time stamps is the first interval. Then, in step S12, the frame data is segmented by the time stamps, so that the three-dimensional point data set is divided into a plurality of first data blocks i according to the first interval.

[0092] Further, the sensor is a wheel sensor, and a plurality of wheel sensors are arranged in sequence on the track; the plurality of wheel sensors are used to generate a set of trigger signals containing the front and rear sequence when the train passes through. The plurality of wheel sensors are arranged in sequence on the track at approximately the first interval, and when the wheels of the train pass through the wheel sensors in sequence, the trigger signals generated by the plurality of wheel sensors contain information of the first interval. The set of trigger signals is recorded by a clock, and a set of time stamps required above can be obtained.

[0093] Preferably, the first interval or the second interval is 300mm to 900mm. When the three-dimensional point data set is a three-dimensional model, the first interval and the second interval can be selected within the above range according to the detection position of the train appearance, requirements, and the operation processing capacity of the equipment. The smaller the first interval and the second interval, the greater the amount of operation in the detection process of meshing and judgment, and the higher the sensitivity or resolution of abnormal situation identification. When the three-dimensional point data set is frame data, the selection of the first interval within the above range can effectively avoid the scanning distortion caused by the fluctuation of the train speed, and further ensure the accuracy of abnormal situation identification from the data source perspective. The speed of the train inevitably fluctuates when passing through the scanning of the laser radar, but due to the inertia of the train and the limitation of the size of acceleration and deceleration, the speed of the train passing through a small range basically remains unchanged. That is, the train driving in a small range can be regarded as uniform motion. Through repeated experimental tests, it can be obtained that the range of the first interval should be between 300mm and 900mm.

[0094] Preferably, the scanning frequency of the laser radar is not less than 500HZ.

[0095] Preferably, the plurality of laser radars are arranged at the upper left corner, the upper right corner, the lower left corner, the lower right corner and the front bottom of the train section respectively, and the scanning area of the laser radar covers the roof and the running part of the train. Among them, the laser radars at the upper left corner and the upper right corner respectively complete the appearance scanning of the train roof from the left and right sides, including the pantograph, refrigeration equipment and other key components; the laser radars at the lower left corner, the lower right corner and the front bottom respectively complete the appearance scanning of the train bottom from the left and right sides and the front, including the bogie and other key components.

[0096] Optionally, a second camera is arranged side by side on one side of the laser radar, and the second camera is configured to synchronously obtain a set of image groups of the appearance of the train when the train passes.

[0097] Optionally, the method further comprises a step S16. If the appearance of the train region corresponding to the second data block ij changes, an alarm message is sent to the user, and an image of the corresponding position is pushed to the user for manual confirmation.

[0098] In this scheme, the second camera is used as a supplement to the laser radar to realize manual auxiliary diagnosis. The second camera is arranged side by side on one side of the laser radar, and the viewing angle range of the second camera covers the train appearance scanned by the laser radar. The second camera can use the above-mentioned timestamp as a mark to confirm the position of the shooting part. When an abnormal situation is identified in the region corresponding to a certain second data block ij, the corresponding timestamp is found through the number i, and the image involving the first data block i region is found from the image group through the timestamp, and finally the image is sent to the user for secondary confirmation of whether an abnormality has occurred.

[0099] Embodiment 2

[0100] As shown in Figure 2 , Figure 9 , Figure 10 The embodiment is a fastener loosening detection method based on 2D image recognition technology, which comprises a first camera and an image recognition module, and the specific steps are as follows:

[0101] S21. A first mark line and a second mark line are arranged on the end face of the fastener; S22. When initialized, the first camera is used to obtain a first image of the end face of the fastener, and the first image is transmitted to the image recognition module; S23. The image recognition module is used to identify the first mark line and the second mark line in the first image, and the reference slope K1 of the first mark line and the reference slope K2 of the second mark line are obtained; S24. When detected, the first camera is used to obtain a second image of the end face of the fastener, and the second image is transmitted to the image recognition module; S25. The image recognition module is used to identify the first mark line and the second mark line in the second image, and the actual slope K5 of the first mark line and the actual slope K6 of the second mark line are obtained; S26. The analysis module is used to judge the slope change of the first mark line and the second mark line, and the loosening result of the fastener is given.

[0102] The scheme takes the fastener loosening detection applied to the bogie of the train as an example. A first camera is fixed on one side of the track to form a certain detection area within the visual angle range of the first camera. A wheel sensor is arranged on the track, and the wheel sensor is used to trigger a shooting signal to the first camera. When the train passes through the detection area at a certain speed, the wheel sensor detects the wheel of the train and sends a shooting signal to the first camera, the first camera shoots an image of the bogie containing a plurality of fasteners, and an image recognition module and an analysis module process the image respectively to obtain the loosening condition of each fastener.

[0103] In the scheme, after the fastener is fastened, a first mark line and a second mark line are arranged on the end face of the fastener by the maintenance personnel. When the train enters the detection area for the first time (at the initialization time), the first camera is triggered to shoot a first image of the bogie, and the first image covers the end face of the fastener; the image recognition module obtains a reference slope K1 of the first mark line and a reference slope K2 of the second mark line. When the train enters the detection area for the second time (at the detection time), the first camera is triggered to shoot a second image of the bogie, and the second image covers the end face of the fastener; the image recognition module obtains an actual slope K5 of the first mark line and an actual slope K6 of the second mark line. Then, the analysis module compares the difference between the reference slope and the actual slope with a preset threshold value, if the difference is greater than the threshold value, it indicates that the fastener is loosened; if the difference is less than the threshold value, it indicates that the fastener is fastened. And there are two ways to calculate the difference between the reference slope and the actual slope, one is to first calculate a first difference between the reference slope K1 and the actual slope K5 of the first mark line, and a second difference between the reference slope K2 and the actual slope K6 of the second mark line, and then average the first difference and the second difference; the other is to first calculate the slope of the angle bisector of the first mark line and the second mark line in the first image as the reference slope, and then calculate the slope of the angle bisector of the first mark line and the second mark line in the second image as the actual slope, and finally calculate the difference between the reference slope and the actual slope of the angle bisector.

[0104] In the scheme, on the one hand, since the first image and the second image obtained by the first camera both include the first mark line and the second mark line, when evaluating the slope change, the analysis module eliminates the error caused by image recognition by solving the average, thereby solving the problem of low accuracy when applying visual detection to fastener loosening. On the other hand, since the double insurance mode of two mark lines is adopted, in the case of image recognition error of one of the mark lines, the analysis module can still determine the loosening result of the fastener through the slope change of the other mark line, thereby solving the problem of high false positive rate when applying visual detection to fastener loosening.

[0105] It should be noted that the first and second entry of the train into the detection area only indicates the order, and does not represent the actual number of entries. That is, when the train enters the detection area next time, the last detection result can be used as initialization data, the last image as the first image, the actual slopes of the first and second mark lines as the reference slopes K1 and K2, and the next slope change analysis is performed. Secondly, the first and second mark lines are not limited to single line segments, but can also be the edges of a figure. For example, the first and second mark lines are two sides of a triangle. Furthermore, according to the viewing angle range of the first camera, the first and second images can cover multiple fasteners, and the image analysis module can number the multiple fasteners according to rules, and the image recognition module and the analysis module can also process the first and second mark lines of multiple fasteners in an image at the same time.

[0106] Preferably, step S23 further comprises: if the image recognition module does not obtain the reference slope K1 of the first mark line and the reference slope K2 of the second mark line, an alarm information is sent to the user, and the step S22 is returned. The added judgment step can ensure that the reference slope K1 of the first mark line and the reference slope K2 of the second mark line are obtained during initialization, so that the loosening of the fastener can be judged by solving the average during detection.

[0107] Preferably, step S27 is further included: if the slope change of the first and second mark lines cannot be identified or exceeds the threshold, an alarm information is sent to the user, and the second image is pushed for manual confirmation at the same time.

[0108] The first and second mark lines are set in the marking manner, and step S21 specifically includes: using a marker to draw the first and second mark lines on the end face of the fastener. Since the maintenance personnel will necessarily set the anti-loosening line after completing the fastening of the fastener, and are required to draw the first and second mark lines, no additional application cost is brought, thereby reducing the influence on the existing safety detection process, making the method easy to popularize and use. In addition, the image recognition module uses a regression algorithm to calculate the slopes of the first and second mark lines, and the first and second mark lines do not need to be straight line segments, thereby reducing the drawing requirements for the maintenance personnel.

[0109] Further, the first mark line and / or the second mark line extends from the end surface of the fastener to the surface of the fastened object. The first mark line or the second mark line on the end surface of the fastener is used for identification and analysis of the recognition module; the first mark line or the second mark line on the extended part is equivalent to the traditional anti-loose line, which is used for manual detection. At this time, it is also equivalent to the maintenance personnel directly drawing the first mark line and the second mark line on the basis of the original anti-loose line. The extended first mark line or second mark line not only adapts to visual detection, ensures the reliability of image recognition, and plays the function of automatic loose detection; but also retains the function of the original anti-loose line, which can be directly compared and checked, realizes the combination of man and machine, and flexible switching between automatic / manual detection.

[0110] Preferably, the first mark line and the second mark line intersect at a point on the end surface of the fastener. After the intersection of the first mark line and the second mark line, the first mark line and the second mark line are associated with each other due to the presence of the intersection point. When the first mark line and the second mark line are identified by the image recognition module, the intersection point is added as an identification feature, which is beneficial to reduce the identification error of the image recognition module and improve the accuracy of the obtained reference slope K1 and reference slope K2.

[0111] In this scheme, the ink of the marker pen should have waterproof properties, reflective properties, and / or waterproof properties and fluorescent properties. The waterproof properties are to prevent the first mark line and the second mark line from being easily damaged, and the reflective properties and fluorescent properties are to improve the exposure of the first mark line and the second mark line in the first image or the second image, increase their contrast, and make the identification of the image recognition module easier and more accurate.

[0112] In this embodiment, the maintenance personnel starts from a point on the edge of the fastener and performs two consecutive straight line markings respectively, extending to the surface of the fastened object to form the first mark line and the second mark line. Among them, the left side is set as the first mark line, and the right side is set as the second mark line, and vice versa. The first mark line and the second mark line can act as traditional anti-loose lines, and the maintenance personnel does not need to draw additional anti-loose lines. The included angle of the first mark line and the second mark line is not less than 30°, and the pixel length of the first mark line and the second mark line in the first image or the second image is not less than 20 pixel points, so as to ensure that the image recognition module can accurately identify and obtain the slope.

[0113] Embodiment 3

[0114] As shown in Figure 2 , Figure 11 , Figure 12 This embodiment is a fastener loosening detection method, which is another implementation based on embodiment 2. Only the differences will be described below, and the same parts will not be repeated.

[0115] The first mark line and the second mark line are arranged in a pasting manner, and step S21 specifically includes: pasting a mark sticker with the first mark line and the second mark line drawn thereon to the end face of the fastener. The first mark line and the second mark line are printed in batches on the mark sticker, and the thickness, straightness, color depth and the like of the first mark line and the second mark line can be uniformly guaranteed, thereby reducing the recognition difficulty of the image recognition module and improving the recognition accuracy. When used, the maintenance personnel can paste the mark sticker on the end face of the fastener substantially in the middle after the fastener is fastened.

[0116] Further, the first mark line and the second mark line are cross lines located at the center of the end face of the fastener. When being cross lines, the included angle of the first mark line and the second mark line is maximum, and the pixel interference between the first mark line and the second mark line is minimum near the intersection point. In addition, the first mark line and the second mark line cover the largest range on the end face of the fastener, and even if the first mark line and the second mark line are locally damaged or locally blocked, the image recognition module is not affected. Thus, the first mark line and the second mark line in the shape of cross lines further improve the recognition accuracy of the image recognition module.

[0117] In the embodiment, the mark sticker is a circular cross reference symbol, including four quadrants, the first quadrant and the third quadrant are dark filled, and the second quadrant and the fourth quadrant are not filled. The image recognition module obtains the first mark line and the second mark line by recognizing the color difference between the quadrants. Among them, the vertical direction is set as the first mark line, and the horizontal direction is set as the second mark line, and vice versa. In addition, the fastener still needs to reserve the anti-loose line to switch to manual detection.

[0118] Further, the dark filled ink in the first quadrant and the third quadrant has waterproof properties, light reflection properties, and / or waterproof properties and fluorescent properties. The waterproof properties are to prevent the first mark line and the second mark line from being easily damaged, and the light reflection properties and the fluorescent properties are to improve the exposure of the first mark line and the second mark line in the first image or the second image, improve the contrast, and make the recognition of the image recognition module easier and more accurate.

[0119] Embodiment 4

[0120] As shown in Figure 13 , Figure 14 , the present embodiment is another implementation of the mark sticker of embodiment 3, and only the differences will be described below, and the same parts will not be repeated.

[0121] The scheme comprises a first mark line, a second mark line and a third mark line, the third mark line intersects with the first mark line and the second mark line respectively; the third mark line is used for correcting the slope change of the first mark line and the second mark line. The increase of the third mark line makes the first mark line and the second mark line increase another intersection point respectively. When the image recognition module identifies the first mark line and the second mark line, two intersection points are added as identification features, which further reduces the identification error of the image recognition module and improves the accuracy of the obtained reference slope K1 and reference slope K2.

[0122] In the embodiment, the mark sticker is a dark-filled triangular sticker. One vertex of the triangle points to the anti-loose line, and along the vertex in the counterclockwise direction, the first mark line, the third mark line and the second mark line can be sequentially arranged. In addition, the ink of the dark-filled triangular sticker should also have waterproof properties, light reflection properties, and / or waterproof properties and fluorescent properties.

[0123] Obviously, the above embodiments of the present application are only examples for clearly illustrating the technical scheme of the present application, and are not intended to limit the specific embodiments of the present application. Any modification, equivalent replacement and improvement made within the spirit and principle of the claims of the present application should be included in the protection scope of the claims of the present application.

Claims

1. A train online detection method, characterized in that, For the region with variable shape position relationship or light transmission characteristics on the train appearance, a third camera is used to obtain a third image of the surface, then an image recognition module is used to identify the features in the third image, and finally an analysis module is used to compare the features of the third images obtained in the first and second times to determine whether the region has changed; for the region with fixed shape position relationship or non-light transmission on the train appearance, a laser radar is used for three-dimensional scanning detection, including the following steps: S11. Scanning the train appearance using the laser radar and obtaining a three-dimensional point data set of the train appearance; S12. Dividing the three-dimensional point data set into a plurality of columns of first data blocks i along the scanning direction at a first interval, i being a positive integer; S13. Dividing each column of first data blocks i into a plurality of rows of second data blocks ij along the vertical direction of the scanning direction at a second interval, i and j being positive integers; S14. Statistically analyzing the distance distribution of three-dimensional points for each second data block ij; S15. A plurality of second data blocks Aij are obtained by first scanning, and a plurality of second data blocks Bij are obtained by second scanning, if the change amplitude of the three-dimensional point distance distribution of the second data block Bij and the second data block Aij is less than a threshold value, then the region of the train appearance corresponding to the second data block ij has no change; Before step S14, the first data block i or the second data block ij is subjected to interpolation or fitting processing of three-dimensional points to make the three-dimensional point density the same.

2. The method of claim 1, wherein, In step S11, the train travels on the track, and the laser radar is fixed on the outside of the track; the train passes through the scanning area of the laser radar, thereby obtaining a three-dimensional point data set of the train appearance.

3. The method of claim 2, wherein, Further comprising a sensor, when the train passes through the scanning area of the laser radar, the sensor is used to trigger a set of time stamp sets containing the time Ti corresponding to each first interval distance of the train advancing, wherein i is a positive integer.

4. The method of claim 3, wherein, The sensor is a wheel sensor, and a plurality of wheel sensors are arranged in sequence on the track; a plurality of wheel sensors are used to generate a set of trigger signals in sequence when the train passes through.

5. The method of claim 2 to 4, wherein, Further comprising a second camera, the second camera is arranged side by side on one side of the laser radar, and the second camera is used to synchronously obtain a set of image sets of the train appearance when the train passes through.

6. The method of claim 1, wherein, The region includes a fastener, and the fastener loosening detection includes the following steps: S21. A first mark line and a second mark line are arranged on the end surface of the fastener; S22. When initializing, a first camera is used to obtain a first image of the end surface of the fastener, and the first image is transmitted to an image recognition module; S23. The image recognition module is used to identify the first mark line and the second mark line in the first image, and the reference slope K1 of the first mark line and the reference slope K2 of the second mark line are obtained; S24. When detecting, a second camera is used to obtain a second image of the end surface of the fastener, and the second image is transmitted to the image recognition module; S25. Identify the first mark line and the second mark line in the second image using the image recognition module, and obtain the actual slope K5 of the first mark line and the actual slope K6 of the second mark line; S26. Use the analysis module to determine the slope change of the first mark line and the second mark line, and give the loosening result of the fastener.

7. The method of claim 6, wherein, Further comprising a third mark line, the third mark line intersects with the first mark line and the second mark line respectively; the third mark line is used to check the slope change of the first mark line and the second mark line.

8. The method of claim 6, wherein, Step S21 specifically comprises: using a marker to draw the first mark line and the second mark line on the end face of the fastener.

9. The method of claim 6 to 8, wherein, Further comprising step S27: if the slope change of the first mark line and the second mark line cannot be identified or exceeds the threshold value, an alarm information is sent to the user, and at the same time, the second image is pushed for manual confirmation.

Images